A novel video noise reduction method based on PDE, adaptive grouping, and thresholding techniques

Abstract

Undoubtedly, video block-matching and 3D filtering (VBM3D) has achieved a significant improvement in video denoising. Nevertheless, in practice, failure to distinguish between the different noise areas, ignoring noise variances and pixel intensity, false-similar patches, and poor matching are the challenges faced by the VBM3D filter. To avoid these drawbacks, a new video denoising algorithm is proposed. This algorithm based on the nature of the noise areas and the spatial distance between the reference block and its candidate blocks. In the algorithm, hard-thresholding in VBM3D is replaced by adaptive filtering. In this adaptive filter, soft-thresholding is applied to the heavily contaminated areas, whereas anisotropic diffusion filter is applied to the slight-noise areas. Applying adaptive filtering creates a balance between noise removal and edges conservation. To avert the occurrence of the poor choice of the threshold, noise variances, clean image coefficients, and pixel intensity are taken into consideration during computing the proposed adaptive threshold. Due to the strong possibility of similar correlative blocks happening in the vicinity, an adaptive grouping technique is proposed to compute the distance between a reference block and its candidate blocks. Applying this technique helps to reduce the occurrence of false-similar blocks and poor matching.